Re: Latch for the WAL writer - further reducing idle wake-ups. - Mailing list pgsql-hackers

From Robert Haas
Subject Re: Latch for the WAL writer - further reducing idle wake-ups.
Date
Msg-id CA+TgmoZ9XgK+H1f1YRXGcx-PfS1Y=sbPFFq_nOxJrEeLrSNb6Q@mail.gmail.com
Whole thread Raw
In response to Re: Latch for the WAL writer - further reducing idle wake-ups.  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
On Wed, May 2, 2012 at 11:42 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> ... It seems unlikely to cause any real
>> problem if WAL writer takes a couple seconds to get with the program
>> after a long period of inactivity; note that an async commit will kick
>> it anyway, and a sync commit will probably half to flush WAL whether
>> the WAL writer wakes up or not.
>
> That's a good point.  What about only kicking the WAL writer in code
> paths where a backend found itself having to write/flush WAL for itself?
> The added overhead is very surely negligible in such a situation.

Yeah, I think that would make sense, though I'd probably still argue
for a hibernation period not quite so long as ten seconds.  Actually,
what I'd really like is for this to be adaptive: if we find that
there's no WAL to write, increase the time until the next wakeup by 10
ms until we hit the maximum of, say, 3 seconds.  If we find that there
is WAL to write, cut the time until the next wakeup in half until we
hit a minimum of, say, 20ms.  And, if we're forced to write/flush WAL
ourselves, or we async commit, kick the WAL writer in the pants and
wake him up right away.  That way we're willing to get
super-aggressive when needed, but we don't stay there very long once
the pounding ends.  Also, we avoid having a hard "cut" between regular
sleeps and deep hibernation; instead, we kind of gradually drift off.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company


pgsql-hackers by date:

Previous
From: Daniel Farina
Date:
Subject: Re: Torn page hazard in ginRedoUpdateMetapage()
Next
From: Tom Lane
Date:
Subject: Re: Torn page hazard in ginRedoUpdateMetapage()